kernel.git/arch/arm64/kernel/stacktrace.c, branch linux-5.16.y

arm64: Mark start_backtrace() notrace and NOKPROBE_SYMBOL

2022-03-08T18:14:04Z

[ Upstream commit 1e0924bd09916fab795fc2a21ec1d148f24299fd ] Mark the start_backtrace() as notrace and NOKPROBE_SYMBOL because this function is called from ftrace and lockdep to get the caller address via return_address(). The lockdep is used in kprobes, it should also be NOKPROBE_SYMBOL. Fixes: b07f3499661c ("arm64: stacktrace: Move start_backtrace() out of the header") Cc: # 5.13.x Signed-off-by: Masami Hiramatsu Reviewed-by: Mark Brown Link: https://lore.kernel.org/r/164301227374.1433152.12808232644267107415.stgit@devnote2 Signed-off-by: Catalin Marinas Signed-off-by: Sasha Levin

arm64: ftrace: use HAVE_FUNCTION_GRAPH_RET_ADDR_PTR

2021-11-16T09:47:54Z

When CONFIG_FUNCTION_GRAPH_TRACER is selected and the function graph tracer is in use, unwind_frame() may erroneously associate a traced function with an incorrect return address. This can happen when starting an unwind from a pt_regs, or when unwinding across an exception boundary. This can be seen when recording with perf while the function graph tracer is in use. For example: | # echo function_graph > /sys/kernel/debug/tracing/current_tracer | # perf record -g -e raw_syscalls:sys_enter:k /bin/true | # perf report ... reports the callchain erroneously as: | el0t_64_sync | el0t_64_sync_handler | el0_svc_common.constprop.0 | perf_callchain | get_perf_callchain | syscall_trace_enter | syscall_trace_enter ... whereas when the function graph tracer is not in use, it reports: | el0t_64_sync | el0t_64_sync_handler | el0_svc | do_el0_svc | el0_svc_common.constprop.0 | syscall_trace_enter | syscall_trace_enter The underlying problem is that ftrace_graph_get_ret_stack() takes an index offset from the most recent entry added to the fgraph return stack. We start an unwind at offset 0, and increment the offset each time we encounter a rewritten return address (i.e. when we see `return_to_handler`). This is broken in two cases: 1) Between creating a pt_regs and starting the unwind, function calls may place entries on the stack, leaving an arbitrary offset which we can only determine by performing a full unwind from the caller of the unwind code (and relying on none of the unwind code being instrumented). This can result in erroneous entries being reported in a backtrace recorded by perf or kfence when the function graph tracer is in use. Currently show_regs() is unaffected as dump_backtrace() performs an initial unwind. 2) When unwinding across an exception boundary (whether continuing an unwind or starting a new unwind from regs), we currently always skip the LR of the interrupted context. Where this was live and contained a rewritten address, we won't consume the corresponding fgraph ret stack entry, leaving subsequent entries off-by-one. This can result in erroneous entries being reported in a backtrace performed by any in-kernel unwinder when that backtrace crosses an exception boundary, with entries after the boundary being reported incorrectly. This includes perf, kfence, show_regs(), panic(), etc. To fix this, we need to be able to uniquely identify each rewritten return address such that we can map this back to the original return address. We can use HAVE_FUNCTION_GRAPH_RET_ADDR_PTR to associate each rewritten return address with a unique location on the stack. As the return address is passed in the LR (and so is not guaranteed a unique location in memory), we use the FP upon entry to the function (i.e. the address of the caller's frame record) as the return address pointer. Any nested call will have a different FP value as the caller must create its own frame record and update FP to point to this. Since ftrace_graph_ret_addr() requires the return address with the PAC stripped, the stripping of the PAC is moved before the fixup of the rewritten address. As we would unconditionally strip the PAC, moving this earlier is not harmful, and we can avoid a redundant strip in the return address fixup code. I've tested this with the perf case above, the ftrace selftests, and a number of ad-hoc unwinder tests. The tests all pass, and I have seen no unexpected behaviour as a result of this change. I've tested with pointer authentication under QEMU TCG where magic-sysrq+l correctly recovers the original return addresses. Note that this doesn't fix the issue of skipping a live LR at an exception boundary, which is a more general problem and requires more substantial rework. Were we to consume the LR in all cases this would result in warnings where the interrupted context's LR contains `return_to_handler`, but the FP has been altered, e.g. | func: | <--- ftrace entry ---> // logs FP & LR, rewrites LR | STP FP, LR, [SP, #-16]! | MOV FP, SP | <--- INTERRUPT ---> ... as ftrace_graph_get_ret_stack() fill not find a matching entry, triggering the WARN_ON_ONCE() in unwind_frame(). Link: https://lore.kernel.org/r/20211025164925.GB2001@C02TD0UTHF1T.local Link: https://lore.kernel.org/r/20211027132529.30027-1-mark.rutland@arm.com Signed-off-by: Mark Rutland Cc: Catalin Marinas Cc: Madhavan T. Venkataraman Cc: Mark Brown Cc: Steven Rostedt Cc: Will Deacon Reviewed-by: Mark Brown Link: https://lore.kernel.org/r/20211029162245.39761-1-mark.rutland@arm.com Signed-off-by: Will Deacon

arm64: Recover kretprobe modified return address in stacktrace

2021-10-22T16:16:53Z

Since the kretprobe replaces the function return address with the kretprobe_trampoline on the stack, stack unwinder shows it instead of the correct return address. This checks whether the next return address is the __kretprobe_trampoline(), and if so, try to find the correct return address from the kretprobe instance list. For this purpose this adds 'kr_cur' loop cursor to memorize the current kretprobe instance. With this fix, now arm64 can enable CONFIG_ARCH_CORRECT_STACKTRACE_ON_KRETPROBE, and pass the kprobe self tests. Signed-off-by: Masami Hiramatsu Acked-by: Will Deacon Signed-off-by: Steven Rostedt (VMware)

arm64: stacktrace: avoid tracing arch_stack_walk()

2021-08-03T09:39:35Z

When the function_graph tracer is in use, arch_stack_walk() may unwind the stack incorrectly, erroneously reporting itself, missing the final entry which is being traced, and reporting all traced entries between these off-by-one from where they should be. When ftrace hooks a function return, the original return address is saved to the fgraph ret_stack, and the return address in the LR (or the function's frame record) is replaced with `return_to_handler`. When arm64's unwinder encounter frames returning to `return_to_handler`, it finds the associated original return address from the fgraph ret stack, assuming the most recent `ret_to_hander` entry on the stack corresponds to the most recent entry in the fgraph ret stack, and so on. When arch_stack_walk() is used to dump the current task's stack, it starts from the caller of arch_stack_walk(). However, arch_stack_walk() can be traced, and so may push an entry on to the fgraph ret stack, leaving the fgraph ret stack offset by one from the expected position. This can be seen when dumping the stack via /proc/self/stack, where enabling the graph tracer results in an unexpected `stack_trace_save_tsk` entry at the start of the trace, and `el0_svc` missing form the end of the trace. This patch fixes this by marking arch_stack_walk() as notrace, as we do for all other functions on the path to ftrace_graph_get_ret_stack(). While a few helper functions are not marked notrace, their calls/returns are balanced, and will have no observable effect when examining the fgraph ret stack. It is possible for an exeption boundary to cause a similar offset if the return address of the interrupted context was in the LR. Fixing those cases will require some more substantial rework, and is left for subsequent patches. Before: | # cat /proc/self/stack | [<0>] proc_pid_stack+0xc4/0x140 | [<0>] proc_single_show+0x6c/0x120 | [<0>] seq_read_iter+0x240/0x4e0 | [<0>] seq_read+0xe8/0x140 | [<0>] vfs_read+0xb8/0x1e4 | [<0>] ksys_read+0x74/0x100 | [<0>] __arm64_sys_read+0x28/0x3c | [<0>] invoke_syscall+0x50/0x120 | [<0>] el0_svc_common.constprop.0+0xc4/0xd4 | [<0>] do_el0_svc+0x30/0x9c | [<0>] el0_svc+0x2c/0x54 | [<0>] el0t_64_sync_handler+0x1a8/0x1b0 | [<0>] el0t_64_sync+0x198/0x19c | # echo function_graph > /sys/kernel/tracing/current_tracer | # cat /proc/self/stack | [<0>] stack_trace_save_tsk+0xa4/0x110 | [<0>] proc_pid_stack+0xc4/0x140 | [<0>] proc_single_show+0x6c/0x120 | [<0>] seq_read_iter+0x240/0x4e0 | [<0>] seq_read+0xe8/0x140 | [<0>] vfs_read+0xb8/0x1e4 | [<0>] ksys_read+0x74/0x100 | [<0>] __arm64_sys_read+0x28/0x3c | [<0>] invoke_syscall+0x50/0x120 | [<0>] el0_svc_common.constprop.0+0xc4/0xd4 | [<0>] do_el0_svc+0x30/0x9c | [<0>] el0t_64_sync_handler+0x1a8/0x1b0 | [<0>] el0t_64_sync+0x198/0x19c After: | # cat /proc/self/stack | [<0>] proc_pid_stack+0xc4/0x140 | [<0>] proc_single_show+0x6c/0x120 | [<0>] seq_read_iter+0x240/0x4e0 | [<0>] seq_read+0xe8/0x140 | [<0>] vfs_read+0xb8/0x1e4 | [<0>] ksys_read+0x74/0x100 | [<0>] __arm64_sys_read+0x28/0x3c | [<0>] invoke_syscall+0x50/0x120 | [<0>] el0_svc_common.constprop.0+0xc4/0xd4 | [<0>] do_el0_svc+0x30/0x9c | [<0>] el0_svc+0x2c/0x54 | [<0>] el0t_64_sync_handler+0x1a8/0x1b0 | [<0>] el0t_64_sync+0x198/0x19c | # echo function_graph > /sys/kernel/tracing/current_tracer | # cat /proc/self/stack | [<0>] proc_pid_stack+0xc4/0x140 | [<0>] proc_single_show+0x6c/0x120 | [<0>] seq_read_iter+0x240/0x4e0 | [<0>] seq_read+0xe8/0x140 | [<0>] vfs_read+0xb8/0x1e4 | [<0>] ksys_read+0x74/0x100 | [<0>] __arm64_sys_read+0x28/0x3c | [<0>] invoke_syscall+0x50/0x120 | [<0>] el0_svc_common.constprop.0+0xc4/0xd4 | [<0>] do_el0_svc+0x30/0x9c | [<0>] el0_svc+0x2c/0x54 | [<0>] el0t_64_sync_handler+0x1a8/0x1b0 | [<0>] el0t_64_sync+0x198/0x19c Cc: Signed-off-by: Mark Rutland Cc: Catalin Marinas Cc: Madhavan T. Venkataraman Cc: Mark Brown Cc: Will Deacon Reviwed-by: Mark Brown Link: https://lore.kernel.org/r/20210802164845.45506-3-mark.rutland@arm.com Signed-off-by: Will Deacon

arm64: stacktrace: use %pSb for backtrace printing

2021-07-08T18:48:22Z

Let's use the new printk format to print the stacktrace entry when printing a backtrace to the kernel logs. This will include any module's build ID[1] in it so that offline/crash debugging can easily locate the debuginfo for a module via something like debuginfod[2]. Link: https://lkml.kernel.org/r/20210511003845.2429846-7-swboyd@chromium.org Link: https://fedoraproject.org/wiki/Releases/FeatureBuildId [1] Link: https://sourceware.org/elfutils/Debuginfod.html [2] Signed-off-by: Stephen Boyd Acked-by: Catalin Marinas Cc: Will Deacon Cc: Jiri Olsa Cc: Alexei Starovoitov Cc: Jessica Yu Cc: Evan Green Cc: Hsin-Yi Wang Cc: Petr Mladek Cc: Steven Rostedt Cc: Andy Shevchenko Cc: Matthew Wilcox Cc: Baoquan He Cc: Borislav Petkov Cc: Dave Young Cc: Ingo Molnar Cc: Konstantin Khlebnikov Cc: Rasmus Villemoes Cc: Sasha Levin Cc: Sergey Senozhatsky Cc: Thomas Gleixner Cc: Vivek Goyal Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds

arm64: stacktrace: Relax frame record alignment requirement to 8 bytes

2021-05-26T19:01:17Z

The AAPCS places no requirements on the alignment of the frame record. In theory it could be placed anywhere, although it seems sensible to require it to be aligned to 8 bytes. With an upcoming enhancement to tag-based KASAN Clang will begin creating frame records located at an address that is only aligned to 8 bytes. Accommodate such frame records in the stack unwinding code. As pointed out by Mark Rutland, the userspace stack unwinding code has the same problem, so fix it there as well. Signed-off-by: Peter Collingbourne Link: https://linux-review.googlesource.com/id/Ia22c375230e67ca055e9e4bb639383567f7ad268 Acked-by: Andrey Konovalov Reviewed-by: Mark Rutland Tested-by: Mark Rutland Link: https://lore.kernel.org/r/20210526174927.2477847-2-pcc@google.com Signed-off-by: Will Deacon

arm64: Change the on_*stack functions to take a size argument

2021-05-26T19:01:17Z

unwind_frame() was previously implicitly checking that the frame record is in bounds of the stack by enforcing that FP is both aligned to 16 and in bounds of the stack. Once the FP alignment requirement is relaxed to 8 this will not be sufficient because it does not account for the case where FP points to 8 bytes before the end of the stack. Make the check explicit by changing the on_*stack functions to take a size argument and adjusting the callers to pass the appropriate sizes. Signed-off-by: Peter Collingbourne Link: https://linux-review.googlesource.com/id/Ib7a3eb3eea41b0687ffaba045ceb2012d077d8b4 Reviewed-by: Mark Rutland Tested-by: Mark Rutland Link: https://lore.kernel.org/r/20210526174927.2477847-1-pcc@google.com Signed-off-by: Will Deacon

arm64: Implement stack trace termination record

2021-05-25T17:53:29Z

Reliable stacktracing requires that we identify when a stacktrace is terminated early. We can do this by ensuring all tasks have a final frame record at a known location on their task stack, and checking that this is the final frame record in the chain. We'd like to use task_pt_regs(task)->stackframe as the final frame record, as this is already setup upon exception entry from EL0. For kernel tasks we need to consistently reserve the pt_regs and point x29 at this, which we can do with small changes to __primary_switched, __secondary_switched, and copy_process(). Since the final frame record must be at a specific location, we must create the final frame record in __primary_switched and __secondary_switched rather than leaving this to start_kernel and secondary_start_kernel. Thus, __primary_switched and __secondary_switched will now show up in stacktraces for the idle tasks. Since the final frame record is now identified by its location rather than by its contents, we identify it at the start of unwind_frame(), before we read any values from it. External debuggers may terminate the stack trace when FP == 0. In the pt_regs->stackframe, the PC is 0 as well. So, stack traces taken in the debugger may print an extra record 0x0 at the end. While this is not pretty, this does not do any harm. This is a small price to pay for having reliable stack trace termination in the kernel. That said, gdb does not show the extra record probably because it uses DWARF and not frame pointers for stack traces. Signed-off-by: Madhavan T. Venkataraman Reviewed-by: Mark Brown [Mark: rebase, use ASM_BUG(), update comments, update commit message] Signed-off-by: Mark Rutland Link: https://lore.kernel.org/r/20210510110026.18061-1-mark.rutland@arm.com Signed-off-by: Will Deacon

Merge tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux

2021-05-07T19:11:05Z

Pull more arm64 updates from Catalin Marinas: "A mix of fixes and clean-ups that turned up too late for the first pull request: - Restore terminal stack frame records. Their previous removal caused traces which cross secondary_start_kernel to terminate one entry too late, with a spurious "0" entry. - Fix boot warning with pseudo-NMI due to the way we manipulate the PMR register. - ACPI fixes: avoid corruption of interrupt mappings on watchdog probe failure (GTDT), prevent unregistering of GIC SGIs. - Force SPARSEMEM_VMEMMAP as the only memory model, it saves with having to test all the other combinations. - Documentation fixes and updates: tagged address ABI exceptions on brk/mmap/mremap(), event stream frequency, update booting requirements on the configuration of traps" * tag 'arm64-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux: arm64: kernel: Update the stale comment arm64: Fix the documented event stream frequency arm64: entry: always set GIC_PRIO_PSR_I_SET during entry arm64: Explicitly document boot requirements for SVE arm64: Explicitly require that FPSIMD instructions do not trap arm64: Relax booting requirements for configuration of traps arm64: cpufeatures: use min and max arm64: stacktrace: restore terminal records arm64/vdso: Discard .note.gnu.property sections in vDSO arm64: doc: Add brk/mmap/mremap() to the Tagged Address ABI Exceptions psci: Remove unneeded semicolon ACPI: irq: Prevent unregistering of GIC SGIs ACPI: GTDT: Don't corrupt interrupt mappings on watchdow probe failure arm64: Show three registers per line arm64: remove HAVE_DEBUG_BUGVERBOSE arm64: alternative: simplify passing alt_region arm64: Force SPARSEMEM_VMEMMAP as the only memory management model arm64: vdso32: drop -no-integrated-as flag

arm64: stacktrace: restore terminal records

2021-04-30T17:30:45Z

We removed the terminal frame records in commit: 6106e1112cc69a36 ("arm64: remove EL0 exception frame record") ... on the assumption that as we no longer used them to find the pt_regs at exception boundaries, they were no longer necessary. However, Leo reports that as an unintended side-effect, this causes traces which cross secondary_start_kernel to terminate one entry too late, with a spurious "0" entry. There are a few ways we could sovle this, but as we're planning to use terminal records for RELIABLE_STACKTRACE, let's revert the logic change for now, keeping the update comments and accounting for the changes in commit: 3c02600144bdb0a1 ("arm64: stacktrace: Report when we reach the end of the stack") This is effectively a partial revert of commit: 6106e1112cc69a36 ("arm64: remove EL0 exception frame record") Signed-off-by: Mark Rutland Fixes: 6106e1112cc6 ("arm64: remove EL0 exception frame record") Reported-by: Leo Yan Tested-by: Leo Yan Cc: Will Deacon Cc: Mark Brown Cc: "Madhavan T. Venkataraman" Link: https://lore.kernel.org/r/20210429104813.GA33550@C02TD0UTHF1T.local Signed-off-by: Catalin Marinas